Review The Intel Core i9-9900KS Special Edition Review: 5.0 GHz on All the Cores, All the Time

[Admin]

Intel's Core i9-9900KS arrives with impressive gaming performance.

The Intel Core i9-9900KS Special Edition Review: 5.0 GHz on All the Cores, All the Time : Read more

 

[Aspiring techie]

If my memory serves me correctly, the Stockfish chess engine got stomped by Google's Deepmind AlphaZero, not the other way around.

 

[atavax]

Why isn't one of the Cons the security vulnerabilities in modern Intel CPUs?

 

[joeblowsmynose]

I just watched Steve Burkes review, and he notes some pretty interesting caveats with this chip ...

The title of this article: "5.0 GHz on All the Cores, All the Time" --- is not true if it is used with a motherboard manufacturer that stuck to Intel's TDP guidelines, like Asus did.

On Asus boards, this chip only boosts all core @5ghz for a limited time then drops back down to maintain reasonable power consumption, as per Intel's own TDP specification for this processor. So basically, Intel gave mobo makers specs to keep TDP ~127w but really it was just their way of lying about TDP but deferring that misinfo to the mobo makers. The mobo maker that actually chose not to allow a misleading TDP now gets punished ... sounds like an Intel move.

So I assume this will mean that Asus is going to be pissed with Intel since gigabyte and MSI boards will let it suck all the power it needs to maintain 5ghz -- completely disregarding the TDP is the only way it boosts at 5ghz all cores, full time.

So how this chip performs has far more to do with the mother board, than the chip. This is stupid.


As an aside question ... what's the cooling power required for OCing? The OC testing here was done using 720mms worth of radiators on a custom loop - what's next ... LN2 testing? We know the limit is somewhere between the H115i and the dual rad custom loop, but I wonder where that is. A lot of cooling for any OCing anyway it seems ... (but expected).

 

[colson79]

I find it sort of annoying how all these reviews always talk about how much better gaming performance is on Intel but leave out the fact that that is only 1080P or lower. The push Intel like it's the only choice for gamers without mentioning the fact that almost every game has the same performance with resolutions over 1080P. I haven't gamed on a 1080 P resolution for years. I think a lot of less informed people completely skip AMD as an option because these review sites push the Intel 1080 P benchmarks so hard. At a minimum I think they should include the 2k and 4k benchmarks in their CPU reviews.

 

[PCWarrior]

I just watched Steve Burkes review, and he notes some pretty interesting caveats with this chip ...On Asus boards, this chip only boosts all core @5ghz for a limited time then drops back down to maintain reasonable power consumption, as per Intel's own TDP specification for this processor.

If "Out of the box" was literally meant to mean “no bios fiddling whatsoever” then stock behaviour should also be with the XMP profile disabled as you need to get into the bios in order to enable it. And for ASUS boards, the moment you go to enable XMP, it prompts you to load optimised defaults which removes power limits. Also it should be pointed out that no-power-limits and MCE are not the same thing and are not viewed as the same thing by Intel. Reviewers like Steve Burkes from Gamers Nexus and Der8auer seem to conflate the two. They are NOT the same. No-power-limits sticks to stock turbo frequency tables (for example the regular 9900K still only boosts to the stock all-core turbo of 4.7 GHZ but instead for only 25 seconds it does so indefinitely). MCE, on the other hand, means both no-power-limits AND to make the all-core turbo boost equal to the single-core turbo boost (for example for the regular 9900K with MCE enabled it means boosting to 5GHZ on all cores indefinitely). For warranty purposes MCE is considered an overclock by Intel. No-power-limits is NOT considered an overclock by Intel. It is stock and it is left to the motherboard vendor how the settings are configured out of the box.

 

[PCWarrior]

I find it sort of annoying how all these reviews always talk about how much better gaming performance is on Intel but leave out the fact that that is only 1080P or lower.

They use 1080p because this is currently the highest resolution where with a top GPU there is definitely a cpu bottleneck and it is therefore a cpu test. It is not a cpu test when there is a gpu bottleneck. With current gpus, even with the likes of 2080Ti, you can have an i3 8100 and still do as well in 4K gaming as with a 9900K. However fast forward to the future and using something like a 3080Ti or a 4080Ti and you will be having a cpu bottleneck across all games at 1440p and probably even at 4K. With a 3080Ti or a 4080Ti you will be getting the same fps that you currently get with a 2080Ti at 1080p but at 1440p and 4K. You can verify this by going backwards to a lower resolution. If you have a 2080Ti, you get the same fps for 720p and 1080p because in a cpu bottleneck situation fps can only increase with a better cpu, not a better gpu.

 

[joeblowsmynose]

If "Out of the box" was literally meant to mean “no bios fiddling whatsoever” then stock behaviour should also be with the XMP profile disabled as you need to get into the bios in order to enable it. And for ASUS boards, the moment you go to enable XMP, it prompts you to load optimised defaults which removes power limits. Also it should be pointed out that no-power-limits and MCE are not the same thing and are not viewed as the same thing by Intel. Reviewers like Steve Burkes from Gamers Nexus and Der8auer seem to conflate the two. They are NOT the same. No-power-limits sticks to stock turbo frequency tables (for example the regular 9900K still only boosts to the stock all-core turbo of 4.7 GHZ but instead for only 25 seconds it does so indefinitely). MCE, on the other hand, means both no-power-limits AND to make the all-core turbo boost equal to the single-core turbo boost (for example for the regular 9900K with MCE enabled it means boosting to 5GHZ on all cores indefinitely). For warranty purposes MCE is considered an overclock by Intel. No-power-limits is NOT considered an overclock by Intel. It is stock and it is left to the motherboard vendor how the settings are configured out of the box.

A complicated way to lie about TDP ... lol. Whatever you say, it disingenuous.

 

[TJ Hooker]

On Asus boards, this chip only boosts all core @5ghz for a limited time then drops back down to maintain reasonable power consumption, as per Intel's own TDP specification for this processor. So basically, Intel gave mobo makers specs to keep TDP ~127w but really it was just their way of lying about TDP but deferring that misinfo to the mobo makers. The mobo maker that actually chose not to allow a misleading TDP now gets punished ... sounds like an Intel move.

FYI this applies to all Intel CPUs with turbo boost, not just the 9900KS. Officially they're all supposed to have a limited duration boost buy nearly all mobos remove this limit. And they've been doing so for some time, so it seems Intel doesn't really care.

Which makes sense, as it improves their benchmark scores and if anyone complains about power draw they can just point to their states rules for power levels and blame the mobo manufacturers, even though they've implicitly given them permission to do this by allowing it to go on in a widespread fashion.

 

[TJ Hooker]

Last page:

Bear in mind that we tested with an Nvidia GeForce GTX 1080 at 1920x1080 to alleviate graphics-imposed bottlenecks.

Should say 2080 Ti.

 

[Soaptrail]

They use 1080p because this is currently the highest resolution where with a top GPU there is definitely a cpu bottleneck and it is therefore a cpu test. It is not a cpu test when there is a gpu bottleneck. With current gpus, even with the likes of 2080Ti, you can have an i3 8100 and still do as well in 4K gaming as with a 9900K. However fast forward to the future and using something like a 3080Ti or a 4080Ti and you will be having a cpu bottleneck across all games at 1440p and probably even at 4K. With a 3080Ti or a 4080Ti you will be getting the same fps that you currently get with a 2080Ti at 1080p but at 1440p and 4K. You can verify this by going backwards to a lower resolution. If you have a 2080Ti, you get the same fps for 720p and 1080p because in a cpu bottleneck situation fps can only increase with a better cpu, not a better gpu.

Yes but they should include a couple 1440p and 4K resolution benchmarks to put context. It could still be beneficial to buyers to get the cheaper options like the AMD Ryzen 3600 and upgrade in a couple years than buy the 9900KS and stick with it for 6 years.

 

[PCWarrior]

FYI this applies to all Intel CPUs with turbo boost, not just the 9900KS. Officially they're all supposed to have a limited duration boost buy nearly all mobos remove this limit. And they've been doing so for some time, so it seems Intel doesn't really care.

Which makes sense, as it improves their benchmark scores and if anyone complains about power draw they can just point to their states rules for power levels and blame the mobo manufacturers, even though they've implicitly given them permission to do this by allowing it to go on in a widespread fashion.

Here are the relevant parts from an interview of Guy Therien, the Intel engineer who is the absolute authority when it comes to Intel’s TDP and turbo.

Guy Therien: Within a processor we define a burst power limit, that we call Power Limit 2 (PL2), compared to the TDP, which is Power Limit 1 (PL1). There is an algorithm that is run in the processor that ensures that TDP, over a period of time, is enforced as the average power. The algorithm isn’t a simple rolling average, but is an exponential weighted moving average […] This PL2 limit, and the power budget, is configurable. In fact, on all client parts, you can configure both the period over which TDP is enforced and the height of the power limit to which you can burst. […]

So if we consider a system at idle, that has a full power budget. As the workload comes onto the cores, the internal weighted time algorithms pump you up to the highest frequency as long as that’s available for the number of cores that are active. Then we’ll look at the power that’s being consumed, and as long as it is less than your power limit PL2, you are good. So, we continue on at that frequency. As the exponential weighted moving average algorithm starts calculating, and sees the power that is being dissipated/sustained. Then, after a time, it’s looking at the time over which it needs to enforce that average power, which we call Tau (τ). When τ is reached, the algorithm kicks in and sees if it needs to reduce the power that’s being consumed, so it can hit can stay within the TDP limit. This means that you get a burst of performance up front. After that performance, that power budget, is exhausted, internally the processor will start to reduce the power by reducing frequency to ensure that the TDP is enforced over the time that it is configured. So, τ is the timeline over which TDP is enforced by the algorithm.

Both PL2 and τ are configurable by the OEM and ODM, and are ‘in spec’. So you have the ability, of course, if the power delivery supports a higher power limit, and also if the system has the thermal capability (because it makes no sense if you’re just going to thermally throttle), to adjust these values. So motherboard manufacturers and ODMs are investing in their power delivery and thermal solutions to allow them to maximize performance or get a certain about of turbo duration without throttling – to the maximum extent possible. As a result of this you can invest different amounts of money into the power delivery, the thermal solution, the thinness of the system and so on – it’s the ability to design something that is differentiated for the audience, both in terms of form factor and performance.[…] So depending on the system capability, we have the TDP, and then the associated Turbo parameters (PL2 and τ) can be configured to meet the capability of the system.

Full interview:

Talking TDP, Turbo and Overclocking: An Interview with Intel Fellow Guy Therien

www.anandtech.com

 

[TJ Hooker]

@PCWarrior I hadn't seen that interview, thanks. Certainly more nuanced than what I said. But I think the takeaway is the same: Turbo behavior (and therefore overall performance) can depend heavily on the BIOS settings.

And it's in the mobo manufacturers best interest to have the default values for those settings as high as possible to get the best 'stock' performance from their products. So we end up in a situation where the TDP value on the CPU spec sheet becomes a somewhat meaningless number for anyone building a PC, because the CPU is free to exceed that power/heat for relatively long periods of time.

 

[PaulAlcorn]

Last page:

Should say 2080 Ti.

Good eye, thanks fixed

 

[nofanneeded]

for me ,

Cons : Price , PCIe LANES and not Gen 3 I can live with Gen 3 with something like 24 lanes at least.

Intel what are you doing you are lost... Such a good CPU needs more lanes ...

with only 16 lanes , the price should be $350 to win over AMD.

 

[splave]

As an aside question ... what's the cooling power required for OCing? The OC testing here was done using 720mms worth of radiators on a custom loop - what's next ... LN2 testing?

Yes 😙😙😙

 

[bit_user]

We put the chip through the wringer in a variety of workloads with different cooling solutions and found that, given adequate cooling, the chip sustains 5.0 GHz on all cores at stock settings,
...
However, Intel's chips shift into Power Level 2 (PL2) during boost activity, which the company defines as 25% over TDP (158.7 Watts for the KS) for a boost duration (Tau) of 28 seconds. Motherboard vendors often ignore boost duration limits, as we see with our own tests above, so you can expect very aggressive boost activity with most motherboards.

@PaulAlcorn , this latter point needs to be emphasized. Not all motherboards will run the CPU at all-core 5.0 GHz sustained, out of the box!

For instance, GamersNexus found that, on the ASUS Maximum XI Hero motherboard, you had to enable the MCE feature, in order for it to sustain PL2, indefinitely.

View: https://www.youtube.com/watch?v=aVLuKqfyVyw

 

[bit_user]

Why isn't one of the Cons the security vulnerabilities in modern Intel CPUs?

Because the mitigations are addressed at the top of page 2, and this CPU has more of them in hardware than any previous model.

Also, on page 2, they say of their testing methodology that:

All of our test results come from the aforementioned operating system and include all publicly available security mitigations.

So, the existing mitigations are already accounted for.

Of course, there could be more, but enough people (including Intel) have been looking for long enough that any new vulnerabilities are probably going to be even more esoteric and even harder to exploit. Probably.

And just because more vulnerabilities have been discovered in Intel CPUs doesn't mean the next one will be.

 

[yeeeeman]

Why isn't one of the Cons the security vulnerabilities in modern Intel CPUs?

Because R0 stepping has mitigations in silicon for them, as Phoronix site found out. Lets stop beating the bush here and move on. No one gives a damn about an average Joe 5000$ economy saving account.

 

[yeeeeman]

A complicated way to lie about TDP ... lol. Whatever you say, it disingenuous.

You can just keep it at 4Ghz and get that 127W TDP. It is up to you.

 

[bit_user]

You can just keep it at 4Ghz and get that 127W TDP. It is up to you.

Well, more attention should be given to the matter, so that users are better able to make informed decisions around what to buy and how they need to configure & cool it.

And speaking of education, one particular myth that should be dispelled is that TDP == baseclocks. What they guarantee is that at TDP, you can sustain at least the base clocks on all cores, but that doesn't mean you can't sustain anything higher - especially on fewer cores.

 

[Raven_BC]

I find it sort of annoying how all these reviews always talk about how much better gaming performance is on Intel but leave out the fact that that is only 1080P or lower. The push Intel like it's the only choice for gamers without mentioning the fact that almost every game has the same performance with resolutions over 1080P. I haven't gamed on a 1080 P resolution for years. I think a lot of less informed people completely skip AMD as an option because these review sites push the Intel 1080 P benchmarks so hard. At a minimum I think they should include the 2k and 4k benchmarks in their CPU reviews.

Bravo! That seems to be far too obvious, not to TH guys, unfortunately. They get too much money from Intel to push their crap and to ignore how Intel destroyed our wallets, Nature, resources, another words our place to live - The Earth.

 

[PCWarrior]

@PaulAlcorn , this latter point needs to be emphasized. Not all motherboards will run the CPU at all-core 5.0 GHz sustained, out of the box!

For instance, GamersNexus found that, on the ASUS Maximum XI Hero motherboard, you had to enable the MCE feature, in order for it to sustain PL2, indefinitely.

On ASUS boards the moment you go to enable the XMP profile you are prompted to “load optimised defaults”. This removes the power limits. But it’s different from enabling multicore enhancement (MCE). If you enable MCE it will indefinitely turbo on all cores to the frequency equal to its single-core turbo, not just to indefinitely turbo to the stock all-core speed (which is the case when you only remove power limits). Now the confusion arises from the fact that the 9900KS has a stock all-core turbo frequency which is equal to its stock single-core turbo frequency so when you remove power limits this is functionally the same as enabling MCE. But you are not really enabling MCE.

On a more philosophical note, what really constitutes “out of the box” behaviour? Just installing the cpu in the motherboard, finishing the PC assembly and turning it on. Not changing anything in the BIOS at all? Then “out of the box” benchmarks should be done with XMP profile disabled, i.e. with DDR4-2133. And no flashing to update your BIOS either. If that were the case all AMD cpus would be getting trashed and the army of AMD fanboys would be upset and be calling the reviewers Intel shills… Here they are even complaining when reviewers test Ryzen cpus with just the maximum officially supported RAM speed instead of the fastest Samsung B-die RAM with the lowest latency (3600MHz C15 or something) the cpu can handle. Or not using the latest AGESA bios update that slightly improves the turbo boost by 25-50MHz. But they are perfectly happy with kneecapping Intel’s performance by taking the high-moral ground of “cpus need to be tested in their out of the box configuration so no changes in the BIOS at all” when it comes to not removing power limits on the Intel side on the ASUS boards...

@PCWarrior I hadn't seen that interview, thanks. Certainly more nuanced than what I said. But I think the takeaway is the same: Turbo behavior (and therefore overall performance) can depend heavily on the BIOS settings. And it's in the mobo manufacturers best interest to have the default values for those settings as high as possible to get the best 'stock' performance from their products. So we end up in a situation where the TDP value on the CPU spec sheet becomes a somewhat meaningless number for anyone building a PC, because the CPU is free to exceed that power/heat for relatively long periods of time.

The point is that in doing so, motherboards vendors are not violating Intel’s spec. It is considered to be perfectly in spec, hence “stock behaviour”. And is covered by Intel's warranty, unlike overclocking. As for TDP becoming meaningless, at least on Intel it is related to actual power consumption. On AMD it is related to “thermal circuit” shenanigans, where TDP equals (temperature difference)/(thermal resistance), and the values of (temperature difference) and (thermal resistance) are simply artificially adjusted so that the quotient gives you whatever TDP you want…

 

[PCWarrior]

Good eye, thanks fixed

There is one more error. In the review you say that the 3900X has 24 PCIE lanes and the 9900KS has 16 lanes. That’s an incorrect comparision as you are counting the 4 lanes dedicated to the chipset lanes on the 3900X’ tally but not on the 9900KS’ tally. I mean if the 3900X is referred to as having a total of 24 lanes then the 9900KS should be referred to as having a total of 20 lanes. And if the 9900KS is referred to as having 16 direct cpu lanes then the 3900X should be referred to as having 20 direct cpu lanes (typically configured as x16 for a graphics card and another x4 for NVMe storage).

 

[TerryLaze]

You can just keep it at 4Ghz and get that 127W TDP. It is up to you.

If you look at the gamers nexus video it keeps 4.4 - 4.6 even on blender using avx...
A far cry from not even being able to reach 4.6 on a single thread ... at all.